Fix cycle detection.

Under some circumstances, trying to find a cycle starting with a given
point can be very time-consuming (probably exponential, as an
implementation of an NP-complete problem), so we lower our expectations,
and just report the first cycle we find, irregardless of which point it
cuts, which is guaranteed to be much faster (quadratic behavior at the
worst--because we won't explore more than a tree out of the graph).

Always find that cycle, even if -q is specified, so that -q is only
`quiet', e.g., does not change the reported result.

Based on a testcase reported by Dragos Ruiu,  okay millert@

0 files changed, 0 insertions, 0 deletions